Applying Machine Learning to Recent Daily Returns

Do recent daily returns for a stock reliably predict its near-term performance? In their January 2026 paper entitled “A Unified Framework for Anomalies based on Daily Returns”, Nusret Cakici, Christian Fieberg, Gabor Neszveda, Robert Bianchi and Adam Zaremba relate the distribution of last-month (21 trading days) daily returns to next-month return without imposing functional forms, via elastic-net regression. They re-estimate the relationship annually using inception-to-date training data. Their approach considers two aspects of past returns:

1. A chronological component that captures the sequence of returns, typically associated with short-term price pressure and liquidity effects.
2. A rank-based component that captures how extreme returns are, commonly linked to behavioral distortions.

They combine these two components into a single Daily Return Information (DRI) signal and compute returns for its corresponding long-short factor, the Daily Return Information Factor (DRIF). The DRIF portfolio is each month long (short) the tenth, or decile, of stocks with the strongest (weakest) expected returns based on DRI. Using monthly firm characteristics and associated daily returns for a broad sample of U.S. stocks, excluding those priced under $5 at end of month and those in the bottom 1% of NYSE market capitalizations, during January 1937 through December 2024, they find that:

(more…)